Page 122 - 南京医科大学学报自然科学版

南京医科大学学报（自然科学版） 第42卷第10期
·1464 · Journal of Nanjing Medical University（Natural Sciences） 2022年10月

·影像研究·

CT纹理特征评估胰腺导管腺癌的分化程度

吴 锦，徐珊珊，张怡帆，汤盛楠，何 健 *
南京大学医学院附属鼓楼医院核医学科，江苏 南京 210008

［摘 要］ 目的：探究基于术前增强纹理特征构建模型对评估胰腺导管腺癌（pancreatic ductal adenocarcinoma，PDAC）分化程
度的价值。方法：回顾性收集2017年1月—2020年10月66例PDAC患者的病例资料，另外34例来自其他医院的PDAC患者被
用于外部验证，根据术后病理结果分为高分化、中⁃低分化两组，分别记录患者的性别、年龄、肿瘤部位、肿瘤最大径、肿瘤强化
程度、血管侵犯情况等临床及常规影像特征，进行单因素回归分析。采用ITK⁃SNAP软件勾画CT检查动、静脉期图像的感兴趣
区（ROI），并提取图像纹理特征。利用单因素分析和二元 Logistic 回归筛选独立预测因子并构建CT纹理特征模型，将训练组
建立的预测模型直接应用于外部验证组，检验模型的准确度。应用受试者工作特征曲线（ROC）的曲线下面积（AUC）评价预测
模型诊断价值。结果：基于动脉期及静脉期分别筛选出 1 个和 2 个纹理特征，分别为运行熵（run entropy）、区域百分比（zone
percentage）和区域大小不均匀性（size⁃zone non⁃uniformity），其成为具有特征性的预测参数并分别构建了预测模型，基于CT动
脉期纹理特征模型在训练组和验证组的AUC、灵敏度及特异度分别为0.716、0.581、0.824和0.722、0.600、0.765；基于CT静脉期
纹理特征模型在训练组和验证组的AUC、灵敏度及特异度分别为0.895、0.781、0.882和0.873、0.722、0.929。结论：CT增强图像
纹理特征在高分化、中⁃低分化PDAC之间存在差异，给术前评估PDAC恶性程度提供了新的方法。
［关键词］ 胰腺导管腺癌；分化程度；纹理分析；鉴别；体层摄影术；X线计算机
［中图分类号］ R735.9 ［文献标志码］ A ［文章编号］ 1007⁃4368（2022）10⁃1464⁃08
doi：10.7655/NYDXBNS20221019

Evaluation of histopathological differentiation in pancreatic ductal adenocarcinoma by
texture analysis of CT

WU Jin，XU Shanshan，ZHANG Yifan，TANG Shengnan，HE Jian *
Department of Nuclear Medicine，Nanjing Drum Tower Hospital，the Affiliated Hospital of Nanjing University Medical
School，Nanjing 210008，China

［Abstract］ Objective：To investigate the value of the texture analysis based on preoperative enhanced CT in predicting the
histopathological differentiation of pancreatic ductal adenocarcinoma（PDAC）. Methods：Pathological data of 66 patients with PDAC
from January 2017 to October 2020 were retrospectively collected，other 34 PDAC patients from other hospital were used for external
validation cohort，and they were divided into two groups according to postoperative pathological results：high differentiation and
moderately ⁃ poorly differentiation. The clinical and conventional imaging characteristics such as gender，age，tumor site，maximum
tumor diameter，tumor enhancement and vascular invasion were recorded for univariate regression analysis. ROI based on both arterial
phase and venous phase of the preoperative enhanced CT was automatically drawn by ITK⁃SNAP software and texture features were
extracted. Univariate analysis was used to compare the texture features between the two groups，and the texture features with statistical
difference were included in binary logistic regression model，and the prediction model of arterial phase and venous phase were
established respectively. The prediction model established by the training group is directly applied to the external validation group. The
AUC values of ROC were used to evaluate the diagnostic value of prediction model. Results：One and two texture features were
selected to construct prediction model respectively based on the CT arterial and venous phase，including run entropy，zone percentage
and size⁃zone non⁃uniformity. The AUC，sensitivity and specificity were 0.716，0.581，0.824，0.722，0.600，and 0.765 in the training
group and the validation group based on the CT arterial texture feature model respectively. The AUC，sensitivity and specificity of the

［基金项目］ 国家自然科学基金（12090023）
∗
通信作者（Corresponding author），E⁃mail：hjxueren@126.com